DeepUbi

DeepUbi predicts lysine ubiquitination sites in proteins to enable identification of ubiquitin-mediated regulation in eukaryotic proteomes.

Key Features:

• Prediction target: Identifies lysine residues modified by ubiquitin as ubiquitination sites in protein sequences.
• Model architecture: Employs convolutional neural networks (CNNs) as the core deep learning model.
• Input features: Utilizes four distinct features derived from protein sequences and their physicochemical properties.
• Data scale: Applicable to large-scale proteome datasets for proteome-wide prediction.
• Evaluation protocol: Performance assessed using 10-fold cross-validation.
• Performance: Reports an AUC (area under the Receiver Operating Characteristic curve) of 0.9, accuracy, sensitivity, and specificity all above 85%, and a Matthews correlation coefficient (MCC) of 0.78.

Scientific Applications:

• Proteome-wide site identification: Enables identification of lysine ubiquitination sites across proteomes.
• Mechanistic studies: Supports investigation of ubiquitin-mediated regulation in signal transduction, cell division, and immune responses.
• Disease-related analysis: Facilitates study of human diseases associated with dysregulated ubiquitin pathways.
• Experimental and clinical research: Assists experimental and clinical studies aimed at elucidating the role of ubiquitination in cellular regulation and disease.

Methodology:

Feature extraction of four sequence- and physicochemical-based descriptors, modeling with convolutional neural networks (CNNs), and performance assessment via 10-fold cross-validation.